Exploring alternative memory architectures for Erlang: Implementation and performance evaluation. Uppsala master thesis in computer science 212

The Erlang/OTP system is currently using a memory architecture with a private heap for each process. This choice was made in an attempt to lower garbage collection times with the copying generational garbage collector the system is using, since the garbage collection stop time interferes with the real-time behavior that Erlang aims at. This however results in costly communication since all messages has to be copied between processes. The aim of this thesis is to design and implement two alternative memory architectures, in an attempt to improve performance of concurrent applications. The rst one is a shared heap memory architecture, where all processes share a global heap. This architecture optimizes inter-process communication as it avoids copying of messages. The shared heap implementation described here is currently shipped together with the Open Source Erlang/OTP release 8 (R8). The other one is an experimental, hybrid architecture: provided that it can be decided at compile time that an allocated object is going to be sent as a message the object is going to be allocated in a global heap. All other objects are going to be placed in the private heap of each process. This hybrid architecture combines the advantages of the private and shared heap architectures. A thorough experimental evaluation of all architectures is also presented. Supervisor and Examiner: Konstantinos Sagonas